Predicting Bitcoin Mining Energy Consumption Using Machine Learning: A Case for k-Nearest Neighbors Regression

Abstract

The energy consumption of Bitcoin mining has emerged as a critical topic in cryptocurrency research, influenced by the significant environmental and economic impacts of blockchain activities. This study examines the energy consumption of Bitcoin mining with a dataset that includes essential blockchain variables such as overall hash rate, network difficulty, daily confirmed transactions, mempool size, average block size, and daily Bitcoin output. A new energy consumption indicator is proposed to contribute to the research domain. The proposed indicator better accurately reflects the dynamics of blockchain energy utilization. Various machine learning models, such as Random Forest, Gradient Boosting, Support Vector Regression, and Multi-layer Perceptron, are evaluated, with particular emphasis on k-Nearest Neighbors Regression (k-NNR). The k-NNR model surpassed all other models, with a 𝑅�2 value of 0.80427 and a Mean Squared Error (MSE) of 0.00441, indicating its high prediction accuracy. Analysis of feature importance indicated that daily Bitcoin production and block size are significant determinants of energy use. The findings underscore the efficacy of k-NNR in energy modeling, offering insights into Bitcoin's energy dynamics and establishing a foundation for more energy-efficient blockchain systems.